Re-pin the vendored cdifTabularData: the counts now carry a propertyID

[smrgeoinfo]

Takes mBB 7052e52a0 (their #48) into vendor/remote/ and re-resolves.

countRows/countColumns were bare unprefixed integers on the tabular distribution — the only two unprefixed properties among twenty-one prefixed siblings in composed adaProduct. Upstream they are now schema:additionalProperty PropertyValue entries whose schema:propertyID names the concept each quantifies: DDI Discovery caseQuantity for rows, variableQuantity for columns.

The vendored diff is only that change

+15 −5 in vendor/remote/.../cdifDataType/cdifTabularData/schema.yaml, and nothing else. Eight other vendored files appear modified in git status with zero content diff — the usual CRLF stat churn. I mis-read that as nine real upstream changes before checking git diff --numstat; worth stating because the vendoring design exists precisely so an unannounced upstream change can't hide in a large diff.

The lock re-pins 49 remote schemas. Downstream: 82 resolvedSchema.json, +10490 −4100, and 0 bare countRows/countColumns declarations remain anywhere in _sources.

The census found a "loss" — and it isn't one

76 blocks each report -1 closed. Cause: two inline copies of the propertyID anyOf — one under prov:wasGeneratedBy, one in $defs/Instrument_AdditionalProperty — collapsed into a single shared $defs/propertyID_item, because the upstream change introduced that as a named $def and the resolver now promotes it once instead of inlining it twice.

Verified the closure survives:

$defs/propertyID_item:
   type=object   required=['@id']   additionalProperties=false
sites now $ref'ing it: 3
   .../prov:wasGeneratedBy/.../schema:propertyID/items      <- was inline
   $defs/Instrument_AdditionalProperty/...                  <- was inline
   $defs/CdifTabularData_AdditionalProperty/...             <- new

One definition serving three sites where two were duplicated — strictly better coverage, one fewer counted node.

This is the census working as intended. 638 examples pass either way, so nothing else in the pipeline would have raised it; the count made me look, and looking showed it benign. Worth recording as a standing characteristic: deduplicating an inline constraint into a shared $def registers as a loss, because the census counts definitions, not the sites reached through them.

Verification

validate_examples	638 passed, 0 failed
validate_counterexamples	13/13 still rejected
constraint_census	247 blocks, 1,355,952 → 1,357,968; baseline updated with --write in this commit

gBB uses these properties in 0 of its own examples, so no example needed rewriting here — the change is schema-only on this side. amds-ldeo/metadata still has two subject-key rows (939 records each) pointing at the old bare path, which is that session's follow-up.

🤖 Generated with Claude Code